Analytics

Analytics is where you find out what actually happened on your calls. Every single call produces a recording, a transcript, and answers to the questions you defined under Post-Call Analysis. This page covers how to read all of it — without getting overwhelmed.

Don't try to look at everything. A successful campaign produces hundreds or thousands of calls. You don't need to listen to all of them. Focus on (1) the headline numbers — completion rate, success rate, cost per outcome — and (2) the 10 best and 10 worst calls. That's where you'll learn what works and what to fix.

Open one call to see the rest. The per-call drill-in view (the sidebar that opens when you click any call) is the most important screen in analytics. Spend time there — audio + transcript + AI analysis side-by-side is how you actually debug.

Where to find analytics

Analytics — deployment execution log with per-call drill-in.

The four metrics that matter

Most users live in these four numbers. Everything else is a debugging tool.

1. Success Rate

Percentage of completed calls that achieved the goal. The goal is whatever you defined in Post-Call Analysis as the primary success field.

Example: if your primary success field is appointment_booked, success rate = (calls with appointment_booked = true) / (calls that connected and completed).

2. Completion Rate

Percentage of attempted calls that reached a real human and got through the conversation (vs. no-answer, busy, voicemail).

Low completion rate = bad list or bad calling hours. Not a Finn problem.

3. Average Duration

How long the average completed call lasts. Useful as a proxy for call quality:

• Too short (<20s) — Finn is being hung up on, or terminating early.
• Too long (>5min) — Finn is rambling, or not closing the loop.
• Steady — pricing predictability.

4. Cost per Successful Outcome

Total cost / count of successful calls. The economic truth. If a campaign costs $0.30 per booked appointment, you can scale it indefinitely.

Per-call view

Click any row in a call table to open the Data Extractor Sidebar.

You'll see:

• Caller info — name, phone, custom fields from the audience.
• Audio player — full recording with waveform scrubbing. Click any point in the transcript to jump to that moment.
• Transcript — speaker-attributed (Finn vs. Caller), timestamped.
• Post-call analysis fields — every structured field you defined, with its extracted value.
• AI analysis — sentiment, summary, outcome category.
• Telemetry — latency, ASR confidence, voicemail detection result, hangup cause.
• Tags — manual or rule-based tags applied to this call.

Navigation

Arrow keys move between calls without closing the sidebar:

• ↑ previous call
• ↓ next call
• Esc close sidebar

Editing transcripts

Click a transcript line to correct it. Edits update the analytics record (and feed back into ASR improvement if you opt in).

Aggregate views

Time series

Daily / hourly trends for success rate, volume, duration. Look for:

• Day-of-week patterns — Tuesdays book; Mondays don't.
• Time-of-day patterns — answer rates spike around lunch.
• Drift — gradual degradation often means your audience is going stale, not that the Finn got worse.

Leaderboards

• Best-performing Finns — sorted by success rate, volume-weighted.
• Worst calls — outlier durations, low sentiment, high latency. Audit these to find Finn bugs.
• High-value contacts — segments that convert above baseline. Feed these back to your CRM.

Post-call analysis breakdown

For every field you defined, a distribution chart:

• Boolean: bar chart of true/false.
• Enum: stacked bar of categories.
• String: word cloud + sample-of-5.
• Number: histogram + median.

Exporting

• Per-deployment CSV — every call row with all fields.
• PDF report — formatted summary suitable for sharing with stakeholders. Includes top-line metrics, charts, and notable transcripts.
• Webhook stream — push every completed call event to your endpoint in real time. Configure under Integrations.

Caching

Analytics aggregates are cached for performance. The dashboard shows a Cached badge when data is from cache. Click Refresh to force a recompute — usually unnecessary unless you suspect staleness.

Cache TTLs:

• Deployment list — 5 minutes
• Per-deployment summary — 1 minute
• Per-call records — never cached (always live)

Real-time vs. final numbers

While a deployment is live, success rate is computed against completed-so-far calls. The number can fluctuate as more calls finish. Wait for the deployment to complete (or for at least 100 calls) before drawing conclusions.

Debugging a bad campaign

If a deployment underperforms, work through this checklist:

1. Open Live Deployments → health panel. Is anything red? Buffer drained? Dispatcher errored?
2. Sort the call table by sentiment ascending. Listen to the worst 5 calls. Patterns are usually obvious within 3 calls.
3. Check completion rate. If completion is low but successful-among-completed is fine, the problem is list quality, not Finn quality.
4. Check post-call analysis fields. Is any field flagged as "extraction failed" disproportionately? That's a transcript or instruction issue.
5. Look at average duration. Way too short = Finn is being hung up on early; check the welcome message.
6. Compare against a known-good Finn. Run the same audience on a baseline Finn for 50 calls and diff the metrics.

Privacy & retention

• Recordings — stored for 90 days by default. Configurable up to 7 years for compliance.
• Transcripts — same retention as recordings.
• PII redaction — opt-in feature in Settings → Compliance. Redacts SSNs, credit card numbers, etc from transcripts and exported data.
• Caller consent — for jurisdictions requiring two-party consent, enable the recording disclaimer in the Finn's welcome message.

Next

• Integrations → — webhooks for real-time analytics streams.
• Billing & Plans → — cost breakdowns and forecasting.
• FAQ → — common analytics questions.